A multi-sensor monitoring system based on internet of things and method thereof

Abstract

This invention discloses a multi-sensor monitoring system and method based on the Internet of Things (IoT), belonging to the field of multi-sensor monitoring technology. It includes: a data acquisition module for collecting patient physiological data and automatically adjusting the sampling strategy according to strategy adjustment instructions; an IoT gateway for receiving and transmitting physiological data and sampling strategy adjustment instructions; a data processing module that, when monitoring a patient post-surgery, sends an abnormal sampling strategy instruction to the IoT gateway, preprocesses the received physiological data, and extracts respiratory, status, and sound features from the preprocessed dataset; and a data visualization module for visualizing monitoring reports. This invention, by integrating multiple physiological features such as respiration, movement, and sound, and comprehensively utilizing the complementarity of different information sources, can more comprehensively and accurately assess physiological status compared to a single data source, reducing the risk of misjudgment.

Description

Technical Field

[0001] This invention relates to the field of multi-sensor monitoring technology, and more specifically to a multi-sensor monitoring system and method based on the Internet of Things. Background Technology

[0002] Currently, patients require monitoring after surgery and during the recovery period. However, current monitoring relies on a single sensor to collect single data. This single-sensor monitoring provides limited information, fails to fully reflect complex physiological states, has poor anti-interference capabilities, and is prone to false alarms / missed alarms, making it impossible to achieve intelligent monitoring strategy adjustments.

[0003] Based on this, the present invention designs a multi-sensor monitoring system and method based on the Internet of Things to solve the above problems. Summary of the Invention

[0004] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a multi-sensor monitoring system and method based on the Internet of Things.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] IoT-based multi-sensor monitoring systems include:

[0007] Data acquisition module: used for collecting patient physiological data, and automatically adjusts the sampling strategy according to strategy adjustment instructions;

[0008] IoT gateway: Connected to the data acquisition module, used to receive and transmit physiological data and sampling strategy adjustment instructions;

[0009] Data Processing Module: Connected to an IoT gateway, when monitoring a patient post-surgery, the module sends an abnormal sampling strategy command to the IoT gateway to preprocess the received physiological data, extracting respiratory, status, and vocal features from the preprocessed dataset. It then fuses these features to determine the current physiological state and generates a monitoring report containing risk level and physiological state. When monitoring a patient in the recovery phase, the module sends a regular sampling strategy command to the IoT gateway to preprocess the received physiological data, extracting respiratory and status features from the preprocessed dataset. It then fuses these features to determine the current physiological state and checks for abnormalities. If an abnormality is detected, the module sends a sampling strategy adjustment command to the IoT gateway and generates a monitoring report containing risk level and physiological state. If an abnormality is detected, the module generates a monitoring report containing risk level and physiological state.

[0010] Data visualization module: Communicates with the data processing module and IoT gateway to visualize monitoring reports and issue alerts. After anomalies in the monitoring report, including risk level and physiological status, are eliminated, manual sampling strategy adjustment instructions can be generated on the operation interface.

[0011] Furthermore, the data acquisition module includes a thin-film sensor, an acceleration sensor, a sound sensor, and a dynamic sampling engine module;

[0012] Thin-film sensor: Integrated under the mattress, used to collect micro-motion data of chest and abdominal pressure waveform.

[0013] Accelerometer sensor: co-located with the thin-film sensor, used for triaxial acceleration motion data;

[0014] Sound sensor: Placed at the head of the bed, used to collect sound data from the audio stream;

[0015] Dynamic sampling engine module: Connects to thin-film sensors, accelerometers, sound sensors, and IoT gateways to control the start and stop of thin-film sensors, accelerometers, and sound sensors according to policy adjustment commands.

[0016] Furthermore, the sampling strategies include routine sampling strategies and abnormal sampling strategies. Routine sampling strategies are suitable for monitoring patients during the recovery period, while abnormal sampling strategies are suitable for patients in the postoperative period or when abnormalities are found in routine sampling strategies.

[0017] The conventional sampling strategy involves real-time monitoring using thin-film sensors and accelerometers.

[0018] The anomaly sampling strategy enables real-time monitoring of the thin-film sensor, accelerometer, and sound sensor.

[0019] Furthermore, the data processing module includes a data judgment module, a preprocessing module, a data extraction module, a data fusion module, and a report generation module;

[0020] Data judgment module: used to determine whether physiological data is postoperative sampling data and to determine whether there are abnormalities in the sampling during the recovery period;

[0021] Preprocessing module: Used for physiological data preprocessing and generating preprocessed datasets;

[0022] Data extraction module: used to extract respiratory features, state features, and sound features from the preprocessed dataset;

[0023] Data fusion module: used to fuse respiratory features, state features and voice features, or respiratory features and state features, and determine the current physiological state based on the fused features;

[0024] Report generation module: Used to generate monitoring reports that include risk level and physiological status based on the current physiological state.

[0025] Furthermore, the data visualization module is also connected to the nurse station communication terminal, the doctor's office communication terminal, and the ward alarm terminal via an IoT gateway.

[0026] A monitoring method employing an Internet of Things (IoT) based multi-sensor monitoring system includes the following steps:

[0027] Step 1: Determine if the patient is in the postoperative period. If yes, proceed to Step 2; if no, proceed to Step 4.

[0028] Step 2: The data acquisition module sends an abnormal sampling strategy instruction to the IoT gateway, and the IoT gateway sends an abnormal sampling strategy instruction to the data acquisition module. The data acquisition module then starts the abnormal sampling strategy to collect the patient's physiological data.

[0029] Step 3: Preprocess the received physiological data to extract respiratory features, state features, and sound features from the preprocessed dataset; then fuse the respiratory features, state features, and sound features to determine the current physiological state based on the fused features, and generate a monitoring report containing risk level and physiological state based on the current physiological state.

[0030] Step 4: The data acquisition module sends a routine sampling strategy instruction to the IoT gateway, and the IoT gateway sends a routine sampling strategy instruction to the data acquisition module. The data acquisition module then starts the routine sampling strategy to collect the patient's physiological data.

[0031] Step 5: Preprocess the physiological data received in Step 4, extract the respiratory features and state features from the preprocessed dataset; then fuse the respiratory features and state features, determine the current physiological state based on the fused features, and determine whether there is an abnormality in the current physiological state. If the determination is yes, execute Step 2 and generate a monitoring report containing the risk level and physiological state. If the determination is no, generate a monitoring report containing the risk level and physiological state.

[0032] Step 6: Receive the monitoring report from Step 3 or Step 5, which includes the risk level and physiological status, display it visually, and provide visual prompts for the risk level.

[0033] Step 7: Determine whether the current sampling strategy is in effect and whether the risk level of the latest monitoring report is "normal". If the determination is yes, proceed to step 8; if the determination is no, proceed to step 2.

[0034] Step 8: When the risk level is "normal" for M consecutive minutes, prompt the medical staff that "the patient's condition is stable and routine monitoring can be resumed." If the medical staff confirms that the condition is normal, proceed to step 4; otherwise, proceed to step 2.

[0035] Furthermore, the specific steps for preprocessing physiological data are as follows:

[0036] Step A: Receive micro-motion data signals, motion data signals, and sound data signals containing start timestamps, and store them in a streaming alignment buffer;

[0037] Step B: Based on the preset common timeline, for the target time point, extract at least two original data points from the streaming alignment buffer whose start timestamps are closest to the target time point;

[0038] Step C: Based on the values ​​of the original data points and the start timestamp, the estimated values ​​of each sensor at the target time point are obtained through interpolation calculation;

[0039] Step D: Combine the estimated values ​​of the sensor at the target time point to form a fused data frame with acquisition time alignment. Use a low-pass filter to filter out high-frequency noise, and then use the moving average method for smoothing. Combine the smoothed micro-motion data frame, smoothed motion data frame and smoothed sound data frame to form a preprocessed dataset.

[0040] Furthermore, the steps for extracting respiratory features are as follows:

[0041] Step 1: Extract respiratory rate features, respiratory rhythm features, and respiratory amplitude features from the fused data frame of aligned micro-motion data using the time domain method;

[0042] Step 2: Remove outliers from respiratory rate, respiratory rhythm, and respiratory amplitude characteristics;

[0043] Step 3: Normalize the respiratory rate, respiratory rhythm, and respiratory amplitude characteristics to remove outliers;

[0044] Step 4: Assemble the normalized respiratory rate features, respiratory rhythm features, and respiratory amplitude features into a one-dimensional respiratory feature in sequence.

[0045] Furthermore, the state feature extraction steps are as follows:

[0046] Step 1: Calculate the resultant acceleration and signal amplitude area from the aligned motion data frames using the time-domain method;

[0047] Step 2: Extract the time-domain features of motion within the resultant acceleration using the time-domain method, and extract the frequency-domain features of motion within the resultant acceleration using Fourier transform;

[0048] Step 3: Extract specific event features using resultant acceleration and signal amplitude area;

[0049] Step 4: Remove outliers from motion time-domain features, motion frequency-domain features, and specific event features;

[0050] Step 5: Normalize the motion time-domain features, motion frequency-domain features, and specific event features that remove outliers;

[0051] Step 6: Assemble the normalized motion time domain features, motion frequency domain features, and specific event features in sequence to form a one-dimensional state feature.

[0052] Furthermore, the specific operations for sound feature extraction are as follows:

[0053] Step 1: The smoothed audio data frames are processed by framing, pre-emphasis, and windowing;

[0054] Step 2: Calculate the discrete cosine transform coefficients and Mel-frequency cepstral coefficients for the windowed signal;

[0055] Step 3: For the discrete cosine transform coefficients along multiple consecutive frames in the time dimension, calculate their first-order and second-order differences to obtain dynamic features and acceleration features.

[0056] Step 4: Input the Mel frequency cepstral coefficients, dynamic features, and acceleration features into a classification model that has been pre-trained on a large amount of labeled audio data to analyze and obtain probability vectors;

[0057] Step 5: Calculate the spectral centroid of the signal after windowing.

[0058] Step 6: Normalize the Mel frequency cepstral coefficients, spectral centroid, and probability vector;

[0059] Step 7: Concatenate the normalized Mel frequency cepstral coefficients, spectral centroid, and probability vector in sequence to form a one-dimensional sound feature.

[0060] Beneficial Effects: This invention integrates multiple physiological characteristics such as respiration, movement, and sound, comprehensively utilizing the complementarity of different information sources. Compared to a single data source, it can more comprehensively and accurately assess physiological status, reducing the risk of misjudgment. It employs a fusion decision-making process based on "respiratory characteristics," "state characteristics," and "sound characteristics" and their dynamic weights. The generated report includes not only the risk level but also a specific description of the "physiological state," enabling medical staff to understand the source of alarms and assisting clinical decision-making. Distinguishing between Postoperative and Rehabilitation Periods: The system can identify patient status and automatically switch monitoring strategies. For high-risk postoperative patients, an "abnormal sampling strategy" including a sound sensor is activated for comprehensive monitoring; for stable patients, a more energy-efficient and less-interference "routine sampling strategy" is used, reducing data processing and storage. During rehabilitation monitoring, once an abnormality is detected, the system automatically upgrades to the "abnormal sampling strategy," achieving risk-based dynamic resource allocation. When the patient's status remains stable, the system prompts medical staff to resume routine monitoring, achieving intelligent closed-loop management of monitoring intensity. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0062] Figure 1 This is a flowchart of the time-aligned multi-sensor monitoring method of the present invention. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0064] The present invention will be further described below with reference to embodiments.

[0065] Example 1: A multi-sensor monitoring system based on the Internet of Things, comprising:

[0066] Data acquisition module: used for collecting patient physiological data, and automatically adjusts the sampling strategy according to strategy adjustment instructions;

[0067] IoT gateway: Connected to the data acquisition module, used to receive and transmit physiological data and sampling strategy adjustment instructions;

[0068] Data Processing Module: Connected to an IoT gateway, when monitoring a patient post-surgery, the module sends an abnormal sampling strategy command to the IoT gateway to preprocess the received physiological data, extracting respiratory, status, and vocal features from the preprocessed dataset. It then fuses these features to determine the current physiological state and generates a monitoring report containing risk level and physiological state. When monitoring a patient in the recovery phase, the module sends a regular sampling strategy command to the IoT gateway to preprocess the received physiological data, extracting respiratory and status features from the preprocessed dataset. It then fuses these features to determine the current physiological state and checks for abnormalities. If an abnormality is detected, the module sends a sampling strategy adjustment command to the IoT gateway and generates a monitoring report containing risk level and physiological state. If an abnormality is detected, the module generates a monitoring report containing risk level and physiological state.

[0069] Data visualization module: Communicates with the data processing module and IoT gateway to visualize monitoring reports and issue alerts. After anomalies in the monitoring report, including risk level and physiological status, are eliminated, manual sampling strategy adjustment instructions can be generated on the operation interface.

[0070] The data acquisition module includes a thin-film sensor, an accelerometer, a sound sensor, and a dynamic sampling engine module;

[0071] Thin-film sensor: Integrated under the mattress, used to collect micro-motion data of chest and abdominal pressure waveform.

[0072] Accelerometer sensor: co-located with the thin-film sensor, used for triaxial acceleration motion data;

[0073] Sound sensor: Placed at the head of the bed, used to collect sound data from the audio stream;

[0074] Dynamic sampling engine module: Connects to thin-film sensors, accelerometers, sound sensors, and IoT gateways to control the start and stop of thin-film sensors, accelerometers, and sound sensors according to policy adjustment commands.

[0075] Sampling strategies include routine sampling strategies and abnormal sampling strategies. Routine sampling strategies are suitable for monitoring patients during the recovery period, while abnormal sampling strategies are suitable for patients in the postoperative period or when abnormalities are found in routine sampling strategies.

[0076] The conventional sampling strategy involves real-time monitoring using thin-film sensors and accelerometers.

[0077] The anomaly sampling strategy involves real-time monitoring using thin-film sensors, accelerometers, and sound sensors.

[0078] The abnormalities found in routine sampling were either abnormal respiratory characteristics or abrupt changes in motion state;

[0079] Physiological data includes micro-motion data, motion data, and sound data;

[0080] The data processing module includes a data judgment module, a preprocessing module, a data extraction module, a data fusion module, and a report generation module;

[0081] Data judgment module: used to determine whether physiological data is postoperative sampling data and to determine whether there are abnormalities in the sampling during the recovery period;

[0082] Preprocessing module: Used for physiological data preprocessing and generating preprocessed datasets;

[0083] Data extraction module: used to extract respiratory features, state features, and sound features from the preprocessed dataset;

[0084] Data fusion module: used to fuse respiratory features, state features and voice features, or respiratory features and state features, and determine the current physiological state based on the fused features;

[0085] Report generation module: Used to generate monitoring reports that include risk level and physiological status based on the current physiological state.

[0086] Example 2, please refer to Figure 1 The monitoring method employs an IoT-based multi-sensor monitoring system, with the following specific steps:

[0087] Step 1: Determine if the patient is in the postoperative period. If yes, proceed to Step 2; if no, proceed to Step 4.

[0088] Step 2: The data acquisition module sends an abnormal sampling strategy instruction to the IoT gateway, and the IoT gateway sends an abnormal sampling strategy instruction to the data acquisition module. The data acquisition module then starts the abnormal sampling strategy to collect the patient's physiological data.

[0089] Step 3: Preprocess the received physiological data to extract respiratory features, state features, and sound features from the preprocessed dataset; then fuse the respiratory features, state features, and sound features to determine the current physiological state based on the fused features, and generate a monitoring report containing risk level and physiological state based on the current physiological state.

[0090] Step 4: The data acquisition module sends a routine sampling strategy instruction to the IoT gateway, and the IoT gateway sends a routine sampling strategy instruction to the data acquisition module. The data acquisition module then starts the routine sampling strategy to collect the patient's physiological data.

[0091] Step 5: Preprocess the physiological data received in Step 4, extract the respiratory features and state features from the preprocessed dataset; then fuse the respiratory features and state features, determine the current physiological state based on the fused features, and determine whether there is an abnormality in the current physiological state. If the determination is yes, execute Step 2 and generate a monitoring report containing the risk level and physiological state. If the determination is no, generate a monitoring report containing the risk level and physiological state.

[0092] Step 6: Receive the monitoring report from Step 3 or Step 5, which includes the risk level and physiological status, display it visually, and provide visual prompts for the risk level.

[0093] Step 7: Determine whether the current sampling strategy is in effect and whether the risk level of the latest monitoring report is "normal". If the determination is yes, proceed to step 8; if the determination is no, proceed to step 2.

[0094] Step 8: When the risk level is "normal" for M consecutive minutes, prompt the medical staff that "the patient's condition is stable and routine monitoring can be resumed." If the medical staff confirms that the condition is normal, proceed to step 4; otherwise, proceed to step 2.

[0095] M is adjusted by medical staff based on the patient's condition;

[0096] When the dynamic sampling engine module starts the thin film sensor, accelerometer sensor, and sound sensor, it adds a unified start-up timestamp to the micro-motion data signal collected by the thin film sensor, the motion data signal collected by the accelerometer sensor, and the sound data signal collected by the sound sensor.

[0097] The specific steps for preprocessing physiological data are as follows:

[0098] Step A: Receive micro-motion data signals, motion data signals, and sound data signals containing start timestamps, and store them in a streaming alignment buffer;

[0099] Step B: Based on the preset common timeline, for the target time point, extract at least two original data points from the streaming alignment buffer whose start timestamps are closest to the target time point;

[0100] Step C: Based on the values ​​of the original data points and the start timestamp, the estimated values ​​of each sensor at the target time point are obtained through interpolation calculation;

[0101] Step D: Combine the estimated values ​​of the sensor at the target time point to form a fused data frame with acquisition time alignment. Use a low-pass filter to filter out high-frequency noise, and then use the moving average method for smoothing. Combine the smoothed micro-motion data frame, smoothed motion data frame and smoothed sound data frame to form a preprocessed dataset.

[0102] The steps for respiratory feature extraction are as follows:

[0103] Step 1: Extract respiratory rate features, respiratory rhythm features, and respiratory amplitude features from the fused data frame of aligned micro-motion data using the time domain method;

[0104] Time-domain method for extracting respiratory rate features: Detect all peaks and troughs, calculate the time interval between adjacent peaks, calculate the average time interval, and calculate the respiratory rate features using the average time interval.

[0105] Respiratory rate characteristic = 60 / average time interval;

[0106] Extracting respiratory rhythm features using the time-domain method: Based on the duration of multiple consecutive respiratory cycles obtained by the time-domain method, calculate the standard deviation of the duration of multiple respiratory cycles, and then divide it by the average duration of multiple respiratory cycles to obtain the respiratory rhythm features;

[0107] Extracting respiratory amplitude features using the time-domain method: Based on the time-domain method, the amplitude difference between the peak and the adjacent trough in each respiratory cycle is calculated, and then the average value is obtained to obtain the respiratory amplitude features;

[0108] Step 2: Remove outliers from respiratory rate, respiratory rhythm, and respiratory amplitude characteristics;

[0109] The outlier removal rules for respiratory rate characteristics, respiratory rhythm characteristics, and respiratory amplitude characteristics are as follows:

[0110] Physiological range filtering: Whether the respiratory rate characteristics are within the set physiologically reasonable range (e.g., adults: 6-40 breaths / min). If it exceeds this range, it is considered as instrument interference or calculation error and will be rejected.

[0111] Statistical outlier detection: For respiratory rate, rhythm, and amplitude sequences calculated over a continuous period of time (e.g., 5 minutes), instantaneous values ​​that significantly deviate from the main distribution are removed using methods based on standard deviation or interquartile range.

[0112] Signal quality index correlation: Combining the quality of micro-motion data signals (such as signal-to-noise ratio), during periods of poor signal quality, even if features are calculated, they are marked as unreliable (abnormal) and removed.

[0113] Step 3: Normalize the respiratory rate, respiratory rhythm, and respiratory amplitude characteristics to remove outliers;

[0114] Step 4: Assemble the normalized respiratory rate features, respiratory rhythm features, and respiratory amplitude features into a one-dimensional respiratory feature in sequence;

[0115] The state feature extraction steps are as follows:

[0116] Step 1: Calculate the resultant acceleration and signal amplitude area from the aligned motion data frames using the time-domain method;

[0117] The resultant acceleration is calculated as follows:

[0118]

[0119] , and Accelerate in the x, y, and z directions;

[0120] Resultant acceleration It can eliminate the influence of sensor orientation and directly reflect the overall motion intensity.

[0121] Calculate the signal amplitude area: Within a sliding time window (e.g., 60 seconds), integrate and sum the absolute values ​​of the triaxial accelerations. The signal amplitude area is a classic indicator for measuring overall activity.

[0122] Step 2: Extract the time-domain features of motion within the resultant acceleration using the time-domain method, and extract the frequency-domain features of motion within the resultant acceleration using Fourier transform;

[0123] Step 3: Extract specific event features using resultant acceleration and signal amplitude area;

[0124] Combined acceleration is the most direct and essential input data for calculating "stationary" and "abrupt" characteristics.

[0125] For the "stationary" characteristic: continuously calculate the variance of the combined acceleration sequence. Only when the variance remains below a threshold (e.g., the variance is less than 0.01g for 5 minutes) will the result be considered. 2 (This allows us to determine the "stationary" state.) Without resultant acceleration, the variance cannot be calculated, and therefore, it is impossible to determine if the object is stationary.

[0126] For the "mutation" feature: A difference operation is performed on the resultant acceleration sequence (i.e., the difference between consecutive data points). The system only marks a "mutation" when the difference value suddenly reaches a maximum value far exceeding the normal range (e.g., greater than 2g). The resultant acceleration is the direct object of the difference operation.

[0127] The combined acceleration is the data source and computational basis for the characteristics of the two events, "rest" and "abrupt".

[0128] Signal amplitude area: Provides "background information" and "aided judgment";

[0129] The relationship between the signal amplitude area and the "static" / "abrupt" features serves as contextual correlation and auxiliary verification.

[0130] Relationship between signal amplitude area and "static" characteristics:

[0131] Logical consistency: When the system determines "stationary" based on the variance of the combined acceleration, the signal amplitude area calculated within the same time window (e.g., 5 minutes) is low (approaching 0). This serves as cross-validation to ensure the reliability of the "stationary" determination.

[0132] The relationship between signal amplitude area and "abrupt change" characteristics:

[0133] Provides an assessment of the event's impact: After a "mutation" event (such as rolling over) occurs, the signal amplitude area will immediately cause a sharp increase in the value of the signal amplitude area within that time window. The signal amplitude area can quantify the energy of this mutation event.

[0134] Distinguishing event types: A violent "rise" (large mutation) and a "slap" (small mutation) can both be detected as mutations, but they leave different imprints on the signal amplitude area, which helps to classify events more finely later.

[0135] The signal amplitude area is not a direct basis for detecting "static" / "abrupt" events, but it provides important context and quantitative supplementation for these events, making the system's judgment more comprehensive and reliable.

[0136] The combined acceleration (derived from variance and difference), signal amplitude area, and the "stationary" / "abrupt" labels together constitute a well-defined and comprehensive feature set:

[0137] Underlying basic characteristics (derived from the resultant acceleration): the resultant acceleration sequence itself, its variance (used to measure rest), and its difference (used to measure abrupt changes).

[0138] Mid-level statistical features (signal amplitude area): The signal amplitude area is the integral statistics of the bottom-level features over a period of time, reflecting the overall activity level.

[0139] High-level semantic features (event labels): "static" and "mutation".

[0140] Specific event characteristics: [signal amplitude area, variance of resultant acceleration, zero-crossing rate of resultant acceleration, low-frequency energy ratio, high-frequency energy ratio, whether stationary (0 / 1), whether abrupt change occurs (0 / 1), intensity of abrupt change];

[0141] Step 4: Remove outliers from motion time-domain features, motion frequency-domain features, and specific event features;

[0142] The types of outliers removed from motion time-domain features, motion frequency-domain features, and specific event features are as follows:

[0143] If the variance of the resultant acceleration is greater than 10 (in g squared), it indicates that the activity is too intense and may be due to abnormal interference. The current time-domain features should be discarded.

[0144] The combined acceleration is greater than 1.5g.

[0145] The peak acceleration at a single sampling point is greater than 3g;

[0146] The frequency centroid is less than 0 or greater than 10Hz.

[0147] The bandwidth is greater than 10Hz.

[0148] Step 5: Normalize the motion time-domain features, motion frequency-domain features, and specific event features that remove outliers;

[0149] Step 6: Assemble the normalized motion time-domain features, motion frequency-domain features, and specific event features in sequence to form a one-dimensional state feature;

[0150] The specific steps for sound feature extraction are as follows:

[0151] Step 1: The smoothed audio data frames are processed by framing, pre-emphasis, and windowing;

[0152] Framing: This process involves cutting a continuous audio signal into overlapping segments (called "frames").

[0153] Frame length: typically 20-40 milliseconds. Too short a frame will result in insufficient frequency resolution, while too long a frame will prevent the signal from being considered stationary.

[0154] Frame shift: usually half the frame length (i.e. 50% overlap);

[0155] For example, a frame length of 30ms and a frame shift of 15ms. Overlapping ensures a smooth transition between frames and prevents the loss of critical information.

[0156] Pre-emphasis operation: Apply a first-order high-pass filter to each frame of signal.

[0157] Objective: To enhance the high-frequency components of a signal to compensate for the natural attenuation of the high-frequency portion of sound during propagation, thereby making the spectrum flatter and facilitating subsequent processing.

[0158] Windowing operation: Multiply the pre-emphasized signal of each frame by a window function (such as a Hamming window).

[0159] Objective: To reduce spectral leakage caused by signal truncation (framing) and to smoothly transition the signals at both ends of the frame to zero.

[0160] Step 2: Calculate the discrete cosine transform coefficients and Mel-frequency cepstral coefficients for the windowed signal;

[0161] The calculation process for the Mel frequency cepstral coefficient is as follows:

[0162] a. Fast Fourier Transform: Converts a windowed signal in the time domain into the frequency domain, obtaining a linear spectrum.

[0163] b. Calculate the power spectrum: Squaring the amplitude of the linear spectrum yields the power spectrum.

[0164] c. Mel-scale filter bank filtering: The power spectrum is passed through a set of Mel-scale triangular filters to obtain the energy value.

[0165] d. Take the logarithm: Take the natural logarithm of the output energy value for each filter.

[0166] e. Discrete Cosine Transform: The energy of the filter bank after taking the logarithm is processed by the discrete cosine transform to obtain the cepstral coefficients.

[0167] f. Retained coefficients: The first 12-13 discrete cosine transform coefficients are retained, and together with the 0th coefficient (representing frame energy), they constitute the Mel frequency cepstral coefficients of a frame.

[0168] Step 3: For the discrete cosine transform coefficients along multiple consecutive frames in the time dimension, calculate their first-order and second-order differences to obtain dynamic features and acceleration features.

[0169] Step 4: Input the Mel frequency cepstral coefficients, dynamic features, and acceleration features into a classification model that has been pre-trained on a large amount of labeled audio data to analyze and obtain probability vectors;

[0170] The probability vector includes [probability of coughing, probability of groaning, probability of normal ambient sound], and the sum is 1;

[0171] The audio types include coughing, groaning, and normal ambient sounds;

[0172] Step 5: Calculate the spectral centroid of the signal after windowing.

[0173] The centroid of the spectrum is calculated as follows:

[0174]

[0175] Step 5: Normalize the Mel frequency cepstral coefficients, spectral centroid, and probability vector;

[0176] Step 6: Concatenate the normalized Mel frequency cepstral coefficients, spectral centroid, and probability vector in sequence to form a one-dimensional sound feature;

[0177] The specific procedures for assessing postoperative fusion characteristics are as follows:

[0178] Step 1: Calculate the quality scores of breathing characteristics, state characteristics, and sound characteristics based on the signal-to-noise ratio, continuity, and stability of the signal;

[0179] Step 2: Sum all the quality scores to get the total score. Divide the quality scores of breathing features, state features, and voice features by the total score to get the weights of breathing features, state features, and voice features respectively.

[0180] Step 3: The weights of the fused features (breathing features, state features, and voice features) are multiplied by each of the breathing features, state features, and voice features, and then added together.

[0181] The specific procedures for assessing postoperative fusion characteristics are as follows:

[0182] Step 1: Calculate the quality scores of breathing characteristics, state characteristics, and sound characteristics based on the signal-to-noise ratio, continuity, and stability of the signal;

[0183] Step 2: Sum all the quality scores to get the total score. Divide the quality scores of breathing features, state features, and voice features by the total score to get the weights of breathing features, state features, and voice features. Obtain the fused features by combining the weights of breathing features, state features, and voice features.

[0184]

[0185]

[0186]

[0187]

[0188]

[0189] , , These are the weights for respiratory features, state features, and vocal features, respectively. , and The quality scores are for respiratory characteristics, state characteristics, and vocal characteristics, respectively. It is the sum of the quality scores of respiratory characteristics, state characteristics, and vocal characteristics.

[0190] The specific procedures for assessing fusion characteristics during the rehabilitation period are as follows:

[0191] Step 1: Calculate the quality scores of respiratory and state characteristics based on the signal-to-noise ratio, continuity, and stability of the signal;

[0192] Step 2: Sum all the quality scores to get the total score. Divide the quality scores of the respiratory feature and the state feature by the total score to get the weights of the respiratory feature and the state feature respectively. Obtain the fused feature by the weights of the respiratory feature and the state feature.

[0193]

[0194]

[0195]

[0196] , These are the weights for respiratory features and state features, respectively. , The quality scores are for respiratory characteristics and state characteristics, respectively. It is the sum of the mass fractions of respiratory characteristics and state characteristics;

[0197] The steps for generating a monitoring report that includes risk level and physiological status are as follows:

[0198] Step 1: Normalize the numerical values ​​of the fused features to a preset mapping model in the 0-1 range, and perform uniform calibration of the numerical range;

[0199] Step 2: Input the calibrated values ​​into the preset mapping model, output a risk score R between 0 and 1, classify the risk score R into risk levels according to the risk assessment criteria, and the preset mapping model outputs a physiological state description based on the risk score R. Based on the classified risk level and the described physiological state, a monitoring report containing the risk level and physiological state is generated.

[0200] The preset mapping model is trained with a large number of clinical samples, which cover physiological characteristic data corresponding to different risk levels.

[0201] If key abnormal features are present (such as respiratory rate <6 breaths / min or >40 breaths / min, peak combined acceleration >3g, groaning probability >0.8), the risk score R is directly corrected to the corresponding high-risk range (such as R≥0.6) to ensure that no critical situation is missed.

[0202] Risk level With risk score Relationship:

[0203]

[0204] The data visualization module is also connected to the nurse station communication terminal, the doctor's office communication terminal, and the ward alarm terminal via an IoT gateway.

[0205] Green: Continue monitoring.

[0206] Yellow: The data visualization module notifies nurses via the nurse station communication terminal.

[0207] Orange: The data visualization module notifies nurses and doctors via the nurse station communication terminal and the attending physician via the doctor's office communication terminal.

[0208] Red: The system notifies nurses via the nurse station communication terminal, notifies attending physicians via the doctor's office communication terminal, and automatically triggers an alarm via the ward alarm terminal.

[0209] This invention integrates multiple physiological characteristics such as respiration, movement, and sound, comprehensively utilizing the complementarity of different information sources. Compared to a single data source, it can more comprehensively and accurately assess physiological status, reducing the risk of misjudgment. It employs a fusion decision-making process based on "respiratory characteristics," "state characteristics," and "voice characteristics" and their dynamic weights. The generated report includes not only the risk level but also a specific description of the "physiological state," enabling medical staff to understand the source of alarms and assisting clinical decision-making. It differentiates between the postoperative and recovery periods: the system can identify patient status and automatically switch monitoring strategies. For high-risk postoperative patients, an "abnormal sampling strategy" including a sound sensor is activated for comprehensive monitoring; for patients in the stable phase, a more energy-efficient and less disruptive "routine sampling strategy" is used, reducing data processing and storage. During recovery monitoring, once an abnormality is detected, the system automatically upgrades to the "abnormal sampling strategy," achieving risk-based dynamic resource allocation. When the patient's status remains stable, the system prompts medical staff to resume routine monitoring, achieving intelligent closed-loop management of monitoring intensity.

[0210] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-sensor monitoring system based on the Internet of Things, characterized in that: include: Data acquisition module: used for collecting patient physiological data, and automatically adjusts the sampling strategy according to strategy adjustment instructions; IoT gateway: Connected to the data acquisition module, used to receive and transmit physiological data and sampling strategy adjustment instructions; Data processing module: Connected to the IoT gateway, when the patient is being monitored post-surgery, it sends an abnormal sampling strategy instruction to the IoT gateway, preprocesses the received physiological data, and extracts respiratory features, state features, and sound features from the preprocessed dataset; The respiratory features, state features, and voice features are then fused together. The current physiological state is determined based on the fused features, and a monitoring report containing risk level and physiological state is generated based on the current physiological state. When the patient is being monitored during the recovery period, a routine sampling strategy instruction is sent to the IoT gateway to preprocess the received physiological data and extract the respiratory features and state features from the preprocessed dataset. The respiratory characteristics and state characteristics are then fused together. The current physiological state is determined based on the fused characteristics. Then, it is determined whether there is an abnormality in the current physiological state. If the determination is yes, a sampling strategy adjustment instruction is sent to the IoT gateway, and a monitoring report containing risk level and physiological state is generated. If the determination is no, a monitoring report containing risk level and physiological state is generated. Data visualization module: Communicates with the data processing module and IoT gateway to visualize monitoring reports and issue alerts. After anomalies in the monitoring report, including risk level and physiological status, are eliminated, manual sampling strategy adjustment instructions can be generated on the operation interface.

2. The multi-sensor monitoring system based on the Internet of Things according to claim 1, characterized in that, The data acquisition module includes a thin-film sensor, an accelerometer, a sound sensor, and a dynamic sampling engine module; Thin-film sensor: Integrated under the mattress, used to collect micro-motion data of chest and abdominal pressure waveforms; Accelerometer sensor: co-located with the thin-film sensor, used for triaxial acceleration motion data; Sound sensor: Placed at the head of the bed, used to collect sound data from the audio stream; Dynamic sampling engine module: Connects to thin-film sensors, accelerometers, sound sensors, and IoT gateways to control the start and stop of thin-film sensors, accelerometers, and sound sensors according to policy adjustment commands.

3. The multi-sensor monitoring system based on the Internet of Things according to claim 2, characterized in that, Sampling strategies include routine sampling strategies and abnormal sampling strategies. Routine sampling strategies are suitable for monitoring patients during the recovery period, while abnormal sampling strategies are suitable for patients in the postoperative period or when abnormalities are found in routine sampling strategies. The conventional sampling strategy involves real-time monitoring using thin-film sensors and accelerometers. The anomaly sampling strategy enables real-time monitoring of the thin-film sensor, accelerometer, and sound sensor.

4. The multi-sensor monitoring system based on the Internet of Things according to claim 1, characterized in that, The data processing module includes a data judgment module, a preprocessing module, a data extraction module, a data fusion module, and a report generation module; Data judgment module: used to determine whether physiological data is postoperative sampling data and to determine whether there are abnormalities in the sampling during the recovery period; Preprocessing module: Used for physiological data preprocessing and generating preprocessed datasets; Data extraction module: used to extract respiratory features, state features, and sound features from the preprocessed dataset; Data fusion module: used to fuse respiratory features, state features and voice features, or respiratory features and state features, and determine the current physiological state based on the fused features; Report generation module: Used to generate monitoring reports that include risk level and physiological status based on the current physiological state.

5. The multi-sensor monitoring system based on the Internet of Things according to claim 1, characterized in that, The data visualization module is also connected to the nurse station communication terminal, the doctor's office communication terminal, and the ward alarm terminal via an IoT gateway.

6. A monitoring method, employing the IoT-based multi-sensor monitoring system as described in claim 1, characterized in that, Includes the following steps: Step 1: Determine if the patient is in the postoperative period. If yes, proceed to Step 2; if no, proceed to Step 4. Step 2: The data acquisition module sends an abnormal sampling strategy instruction to the IoT gateway, and the IoT gateway sends an abnormal sampling strategy instruction to the data acquisition module. The data acquisition module then starts the abnormal sampling strategy to collect the patient's physiological data. Step 3: Preprocess the received physiological data and extract respiratory features, state features, and sound features from the preprocessed dataset; The respiratory characteristics, state characteristics, and voice characteristics are then fused together. The current physiological state is determined based on the fused characteristics, and a monitoring report containing the risk level and physiological state is generated based on the current physiological state. Step 4: The data acquisition module sends a routine sampling strategy instruction to the IoT gateway, and the IoT gateway sends a routine sampling strategy instruction to the data acquisition module. The data acquisition module then starts the routine sampling strategy to collect the patient's physiological data. Step 5: Preprocess the physiological data received in Step 4, and extract respiratory and state features from the preprocessed dataset; Then, the respiratory characteristics and state characteristics are fused together, and the current physiological state is determined based on the fused characteristics. It is then determined whether there is an abnormality in the current physiological state. If the determination is yes, step 2 is executed, and a monitoring report containing the risk level and physiological state is generated. If the determination is no, a monitoring report containing the risk level and physiological state is generated. Step 6: Receive the monitoring report from Step 3 or Step 5, which includes the risk level and physiological status, display it visually, and provide visual prompts for the risk level. Step 7: Determine whether the current sampling strategy is in effect and whether the risk level of the latest monitoring report is "normal". If the determination is yes, proceed to step 8; if the determination is no, proceed to step 2. Step 8: When the risk level is "normal" for M consecutive minutes, prompt the medical staff that "the patient's condition is stable and routine monitoring can be resumed." If the medical staff confirms that the condition is normal, proceed to step 4; otherwise, proceed to step 2.

7. The monitoring method according to claim 6, characterized in that, The specific steps for preprocessing physiological data are as follows: Step A: Receive micro-motion data signals, motion data signals, and sound data signals containing start timestamps, and store them in a streaming alignment buffer; Step B: Based on the preset common timeline, for the target time point, extract at least two original data points from the streaming alignment buffer whose start timestamps are closest to the target time point; Step C: Based on the values ​​of the original data points and the start timestamp, the estimated values ​​of each sensor at the target time point are obtained through interpolation calculation; Step D: Combine the estimated values ​​of the sensor at the target time point to form a fused data frame with acquisition time alignment. Use a low-pass filter to filter out high-frequency noise, and then use the moving average method for smoothing. Combine the smoothed micro-motion data frame, smoothed motion data frame and smoothed sound data frame to form a preprocessed dataset.

8. The monitoring method according to claim 7, characterized in that, The steps for respiratory feature extraction are as follows: Step 1: Extract respiratory rate features, respiratory rhythm features, and respiratory amplitude features from the fused data frame of aligned micro-motion data using the time domain method; Step 2: Remove outliers from respiratory rate, respiratory rhythm, and respiratory amplitude characteristics; Step 3: Normalize the respiratory rate, respiratory rhythm, and respiratory amplitude characteristics to remove outliers; Step 4: Assemble the normalized respiratory rate features, respiratory rhythm features, and respiratory amplitude features into a one-dimensional respiratory feature in sequence.

9. The monitoring method according to claim 7, characterized in that, The state feature extraction steps are as follows: Step 1: Calculate the resultant acceleration and signal amplitude area from the aligned motion data frames using the time-domain method; Step 2: Extract the time-domain features of motion within the resultant acceleration using the time-domain method, and extract the frequency-domain features of motion within the resultant acceleration using Fourier transform; Step 3: Extract specific event features using resultant acceleration and signal amplitude area; Step 4: Remove outliers from motion time-domain features, motion frequency-domain features, and specific event features; Step 5: Normalize the motion time-domain features, motion frequency-domain features, and specific event features that remove outliers; Step 6: Assemble the normalized motion time domain features, motion frequency domain features, and specific event features in sequence to form a one-dimensional state feature.

10. The monitoring method according to claim 7, characterized in that, The specific steps for sound feature extraction are as follows: Step 1: The smoothed audio data frames are processed by framing, pre-emphasis, and windowing; Step 2: Calculate the discrete cosine transform coefficients and Mel-frequency cepstral coefficients for the windowed signal; Step 3: For the discrete cosine transform coefficients along multiple consecutive frames in the time dimension, calculate their first-order and second-order differences to obtain dynamic features and acceleration features. Step 4: Input the Mel frequency cepstral coefficients, dynamic features, and acceleration features into a classification model that has been pre-trained on a large amount of labeled audio data to analyze and obtain probability vectors; Step 5: Calculate the spectral centroid of the signal after windowing. Step 6: Normalize the Mel frequency cepstral coefficients, spectral centroid, and probability vector; Step 7: Concatenate the normalized Mel frequency cepstral coefficients, spectral centroid, and probability vector in sequence to form a one-dimensional sound feature.

Images